Gradient
An R&D lab building decentralized AI infrastructure
Performances
RD Popularity Index
RD Growth Index
Comparison
Follow Updates
Follow Lists
People
About Gradient
Gradient is a Solana-based distributed AI infrastructure project that moves model inference and post-training onto an open network of ordinary devices rather than a handful of cloud data centres. Its stack has three parts: Lattica carries weights and inference traffic between nodes, Parallax splits a model across a pool of heterogeneous GPUs and runs it cooperatively, open-sourced in September 2025, and Echo distributes reinforcement-learning post-training. Scheduling allocates shards by compute and bandwidth, using pipeline parallelism and speculative decoding to hold latency down.
Its users are developers who want cheaper inference and research teams without GPU quotas: Gradient Cloud resells that distributed capacity behind an OpenAI-compatible API, and a chat front end serves as the public demo for Parallax. The node network accumulated earlier through a browser extension has become the communication substrate for Lattica, the focus has shifted to open frameworks and serving, and capacity is scheduled in tiers by device class.
Gradient Network, an open AI infrastructure layer, has marked several milestones with long-term significance. It secured a $10 million seed round led by Pantera Capital and Multicoin Capital, validating its vision of decentralized AI. A subsequent funding round with Multicoin participation further strengthened its position. The project also gained notable traction, becoming the most-followed new project on RootData within 7 days, reflecting growing community interest. These events underscore its progress toward building a peer-to-peer compute and data layer for AI, moving beyond centralized models.
On 11 February 2026, Gradient released Echo-2, a distributed reinforcement learning system that splits AI post-training across heterogeneous devices. Company benchmarks showed the cost of training a 30 billion parameter model falling from about USD 4,490 on commercial platforms to USD 425, a reduction above 90 percent, with post-training costs up to 80 percent lower than conventional cloud approaches.
On the product side, Gradient has shifted from its early Sentry Node browser extension toward engineering-heavy infrastructure: Parallax is open source on GitHub and supports distributed inference, Lattica moves model weights and inference data between nodes, and Gradient Cloud offers an inference API to developers. Official documentation marks the Sentry Node beta as concluded and Edge Host as invite-only, with no token or generation event announced.
Future strategy and roadmap: Gradient plans to broaden Parallax model support and Echo training scale in the second half of 2026, connect consumer device compute to more enterprise inference workloads, and define tokenomics and incentives only after the product and network mature.
Eric Yang — Founder & CEO; formerly at Sequoia Capital China and previously co-founded and exited a Web3 livestreaming company.
Yuan Gao — Co-founder; led early growth at the Helium Network.
Alex Mirran — Business Development lead; came up through the Filecoin and IPFS communities.
Joy Song — Head of Marketing.
The team previously built a decentralized CDN reaching about 900,000 nodes; founded in 2024, with a $10M seed round from Pantera, Multicoin, Sequoia and others.
According to linkedin.com, Alex Mirran, Head of Business Development for North America at Gradient, left the company in July 2026. He joined in September 2025 and served for about 10 months, leading North American GTM sales and partnerships, scaling Commonstack.ai from 0 to 6,500 users.
After leaving, he co-founded GridWeave in August 2026 and became its COO, while continuing as a research consultant at VenHedge Asset Management. Gradient has not yet announced a successor; its open-source tools and commercial products continue to advance.